Intern 曾順承
description
Transcript of Intern 曾順承
![Page 1: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/1.jpg)
Intern Intern 曾順承曾順承
![Page 2: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/2.jpg)
IntroductionIntroduction11
Computerized medical decision suppoComputerized medical decision support systemsrt systems Discriminant analysis Discriminant analysis Logistic regression Logistic regression Recursive partitioningRecursive partitioning Neural networksNeural networks
![Page 3: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/3.jpg)
IntroductionIntroduction22
This report compares three different This report compares three different mathematical models for building a trmathematical models for building a traumatic brain injury (TBI) medical decaumatic brain injury (TBI) medical decision support system (MDSS)ision support system (MDSS) A logistic regression model A logistic regression model A multi-layer perceptron (MLP) neural netA multi-layer perceptron (MLP) neural net
workwork A radial-basis-function (RBF) neural netwA radial-basis-function (RBF) neural netw
orkork
![Page 4: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/4.jpg)
IntroductionIntroduction33
![Page 5: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/5.jpg)
IntroductionIntroduction44
![Page 6: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/6.jpg)
IntroductionIntroduction55
These models were developed based These models were developed based on a large TBI patient database. on a large TBI patient database.
This MDSS accepts a set of patient This MDSS accepts a set of patient datadata The types of skull fracture The types of skull fracture Glasgow Coma Scale (GCS) Glasgow Coma Scale (GCS) Episode of convulsion Episode of convulsion Return the chance that a neurosurgeon Return the chance that a neurosurgeon
would recommend an open-skull would recommend an open-skull surgery for this patient.surgery for this patient.
![Page 7: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/7.jpg)
MethodMethod11
From the 12640 patients selected From the 12640 patients selected from the databasefrom the database A randomly drawn 9480 cases were A randomly drawn 9480 cases were
used as the training group to used as the training group to develop/train our modelsdevelop/train our models
The other 3160 cases were in the The other 3160 cases were in the validation group which we used to validation group which we used to evaluate the performance of these evaluate the performance of these modelsmodels
![Page 8: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/8.jpg)
MethodMethod22
The indicator of how accurate these The indicator of how accurate these models are in predicting a models are in predicting a neurosurgeon’s decision on open-neurosurgeon’s decision on open-skull surgeryskull surgery Sensitivity Sensitivity Specificity Specificity Areas under receiver-operating Areas under receiver-operating
characteristics (ROC) curve and characteristics (ROC) curve and calibration curvescalibration curves
![Page 9: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/9.jpg)
ResultResult11
Assuming equal importance of Assuming equal importance of sensitivity and specificity sensitivity and specificity The logistic regression model The logistic regression model
(sensitivity, specificity) of(73%, 68%)(sensitivity, specificity) of(73%, 68%) The RBF model (80%, 80%) The RBF model (80%, 80%) The MLP model (88%, 80%) The MLP model (88%, 80%)
![Page 10: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/10.jpg)
ResultResult22
![Page 11: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/11.jpg)
ResultResult33
The resultant areas under ROC The resultant areas under ROC curvecurve Logistic regression: 0.761 Logistic regression: 0.761 RBF: 0.880RBF: 0.880 MLP: 0.897 MLP: 0.897 P < 0.05 P < 0.05
Among these models, the logistic Among these models, the logistic regression has noticeably poorer regression has noticeably poorer calibration.calibration.
![Page 12: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/12.jpg)
ResultResult44
![Page 13: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/13.jpg)
ConclusionConclusion11
This study demonstrated the This study demonstrated the feasibility of applying neural feasibility of applying neural networks as the mechanism for TBI networks as the mechanism for TBI decision support systems based on decision support systems based on clinical databases. clinical databases.
![Page 14: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/14.jpg)
ConclusionConclusion22
The results also suggest that neural The results also suggest that neural networks may be a better solution networks may be a better solution for complex, non-linear medical for complex, non-linear medical decision support systems than decision support systems than conventional statistical techniques conventional statistical techniques such as logistic regression.such as logistic regression.
![Page 15: Intern 曾順承](https://reader036.fdocuments.net/reader036/viewer/2022081603/56814c09550346895db90a10/html5/thumbnails/15.jpg)
Thank you for Thank you for attention!attention!